what's new in Wonder3D++?

[adkAurora]

No description provided.

[YuxiaoYang23]

Hi there,

We’ve made some exciting updates:

1. the addition of a camera type switcher, allowing the model to handle input images captured with various camera settings;
2. the implementation of multi-domain pre-training, ensuring a seamless transition from stable diffusion pre-training to the desired joint distribution for 3D data; and
3. replacing implicit neural SDF fields with explicit mesh representations for 3D textured mesh reconstruction, introducing a cascaded 3D mesh extraction algorithm that greatly improves both efficiency and effectiveness.

Additionally, we’ve updated the paper in the repository—please feel free to refer to it for more details.